|
Post by Bones on Jan 2, 2022 11:06:39 GMT -5
This looks like it's going to be a good comp, spread the word guys and let's see how competitive it can really be.
|
|
|
Post by mllrkllr88 on Jan 2, 2022 11:58:23 GMT -5
For real though. Look at this lol. I mean CBR15 isn't exactly the heaviest thing out there. <button disabled="" class="c-attachment-insert--linked o-btn--sm">Attachment Deleted</button> Shocked you got even that far...
|
|
|
Post by PolRoger on Jan 2, 2022 12:22:19 GMT -5
32nm is the way to go here, even the best bloomfields out there will have a hard time reach 5.2 on ambient cooling. On cold same story, best bloomfields will top out around 5600-5700 no matter the bench, while westmere can keep trucking at 6ghz and beyond. I think I'll see what my backup chips can do, don't wanna degrade the best I have pushing on ambient, only have chiller here, so don't think I'll run proper cold for now. Good idea!... Do you cool the top of your air cooler with your phase block/head while benching? I sometimes put a rad in an ice bucket to lower my water loop temps but I don't think I've ever seen that before. I've also run an air-cooled setup outside during a cold weather snap when the ambient temps drop below freezing.
|
|
|
Post by georgekokovinis on Jan 2, 2022 12:39:23 GMT -5
You are funny guys. And I will take the plane for Siberia and bench at -40c. But I will consider ambient.
Please dont do here the crap happening at HWBOT for 20 years now.
In case you dont understand ambient I can explain.
|
|
|
Post by ShrimpBrime on Jan 2, 2022 12:50:42 GMT -5
You are funny guys. And I will take the plane for Siberia and bench at -40c. But I will consider ambient. Please dont do here the crap happening at HWBOT for 20 years now. In case you dont understand ambient I can explain. I have a reading of 50F in my work space. See anything funny George, we will ask the screen shots to be taken with core temp, ambient and loaded temps if need be. I'm pretty aware that these chips don't do much past 4.6ghz with any real load and v-core on ambient. 4.8ghz without HT and core reduction would be questionable. (Golden samples excluded I guess.....??) But I'm sure we'll be fine
|
|
|
Post by ShrimpBrime on Jan 2, 2022 12:51:54 GMT -5
For real though. Look at this lol. I mean CBR15 isn't exactly the heaviest thing out there. <button disabled="" class="c-attachment-insert--linked o-btn--sm">Attachment Deleted</button> Shocked you got even that far... I can do amazing things sometimes. lol
|
|
|
Post by ground on Jan 2, 2022 13:39:05 GMT -5
32nm is the way to go here, even the best bloomfields out there will have a hard time reach 5.2 on ambient cooling. On cold same story, best bloomfields will top out around 5600-5700 no matter the bench, while westmere can keep trucking at 6ghz and beyond. I think I'll see what my backup chips can do, don't wanna degrade the best I have pushing on ambient, only have chiller here, so don't think I'll run proper cold for now. <button disabled="" class="c-attachment-insert--linked o-btn--sm">Attachment Deleted</button> Good idea!... Do you cool the top of your air cooler with your phase block/head while benching? I sometimes put a rad in an ice bucket to lower my water loop temps but I don't think I've ever seen that before. I've also run an air-cooled setup outside during a cold weather snap when the ambient temps drop below freezing. Using an empty CPU pot to weigh down the air cooler Keep in mind I'm using crazy binned CPUs here with no mind for risk of degradation - they are BCLK limited on cold anyways and I have better BCLK chips of this kind anyways. Can happily record a run of pifast at silly clocks with rig in video or with coretemp open
|
|
|
Post by PolRoger on Jan 2, 2022 13:43:01 GMT -5
You are funny guys. And I will take the plane for Siberia and bench at -40c. But I will consider ambient. Please dont do here the crap happening at HWBOT for 20 years now. In case you dont understand ambient I can explain. I have a reading of 50F in my work space. See anything funny George, we will ask the screen shots to be taken with core temp, ambient and loaded temps if need be. I'm pretty aware that these chips don't do much past 4.6ghz with any real load and v-core on ambient. 4.8ghz without HT and core reduction would be questionable. (Golden samples excluded I guess.....??) But I'm sure we'll be fine My basement ambient temps now are at ~78F (Atlanta). I'm currently testing with and an air cooler at ~4.6GHz. I'm planning on switching over to a custom water-cooling loop. I may even try at some point with a rad in an ice bucket but there is no glycol in my loop so even with an ice bath the water temps will be above freezing.
|
|
|
Post by ShrimpBrime on Jan 2, 2022 13:56:43 GMT -5
I have a reading of 50F in my work space. See anything funny George, we will ask the screen shots to be taken with core temp, ambient and loaded temps if need be. I'm pretty aware that these chips don't do much past 4.6ghz with any real load and v-core on ambient. 4.8ghz without HT and core reduction would be questionable. (Golden samples excluded I guess.....??) But I'm sure we'll be fine My basement ambient temps now are at ~78F (Atlanta). I'm currently testing with and an air cooler at ~4.6GHz. I'm planning on switching over to a custom water-cooling loop. I may even try at some point with a rad in an ice bucket but there is no glycol in my loop so even with an ice bath the water temps will be above freezing. <button disabled="" class="c-attachment-insert--linked o-btn--sm">Attachment Deleted</button> 4.6ghz sounds right. All core, that's probably about your max I'd wager. I'm all for chilling in the Ambient stage, just remember those Cpu temps need to be above freezing, we will ask for proof even from admin and myself. Or you guys can just grab coretemp, cover your butts so you don't need to be asked later..... It's all common sense I think. Although I trust most everyone, so I doubt there will be any issues.
|
|
|
Post by ShrimpBrime on Jan 2, 2022 15:49:39 GMT -5
HAHA!!!!!
Finally figured out how to post the 2133mhz strap. OMG I feel like it's going somewhere now. lol. Honestly though, bus clocking with mixed memory brings some fast instability.....
Question, I'm pretty sure already, but There seems not to be an option to raise QPI above "Auto" which sets the highest multiplier any ways. Is this correct. Only bus clocks will increase QPI and Uncore frequencies correct?
|
|
|
Post by mllrkllr88 on Jan 2, 2022 16:14:57 GMT -5
I want to see some efficiency comparisons What shall the terms be? I suggest 32M with a 4G CPU limit. We can have a side competition to see how the efficiency lines up. I will be running PSC and I am curious how far off it is from Hyper...I suspect pretty small difference.
|
|
|
Post by ShrimpBrime on Jan 2, 2022 16:19:42 GMT -5
I want to see some efficiency comparisons What shall the terms be? I suggest 32M with a 4G CPU limit. We can have a side competition to see how the efficiency lines up. I will be running PSC and I am curious how far off it is from Hyper...I suspect pretty small difference. Oh I'm far from having this tuned. I'm at 10-11-10 1.65v. I'm comfortable with that. 17.92 PiFast about 13 seconds ago I just ran it. Full windows with Interwebs. Figured I'd practice and learn this hardware way before I submit anything. A side comp I can do. Let me hit go on a 32m after I close the browser. 136.48 x 35 @ 4777mhz is what I'm running right now. (I don't know if it will pass a Pi 32m run, I haven't tried that yet. - I have absolutely no clue where this chip stops and starts, it's the first time running it along with the board.....)
|
|
|
Post by Mr.Scott on Jan 2, 2022 16:26:56 GMT -5
I want to see some efficiency comparisons What shall the terms be? I suggest 32M with a 4G CPU limit. We can have a side competition to see how the efficiency lines up. I will be running PSC and I am curious how far off it is from Hyper...I suspect pretty small difference. Very small difference. I'll give this a go on the side. I'm no 32m god though. I just hold my own.
|
|
|
Post by PolRoger on Jan 2, 2022 16:28:21 GMT -5
Question, I'm pretty sure already, but There seems not to be an option to raise QPI above "Auto" which sets the highest multiplier any ways. Is this correct. Only bus clocks will increase QPI and Uncore frequencies correct? There are 4 QPI speed settings... Slow mode; 4.8GT/s=18x; 5.866GT/s= 22x; and 6.400GT/s = 24x. Different motherboard BIOS have different ways of reading QPI but CPUZ will read it in MHz. I have found that ~4000MHz QPI speed seems to be the upper limit in most cases.
|
|
|
Post by ShrimpBrime on Jan 2, 2022 16:33:25 GMT -5
Question, I'm pretty sure already, but There seems not to be an option to raise QPI above "Auto" which sets the highest multiplier any ways. Is this correct. Only bus clocks will increase QPI and Uncore frequencies correct? There are 4 QPI speed settings... Slow mode; 4.8GT/s=18x; 5.866GT/s= 22x; and 6.400GT/s = 24x. Different motherboard BIOS have different ways of reading QPI but CPUZ will read it in MHz. I have found that ~4000MHz QPI speed seems to be the upper limit in most cases. Ok, I'm running 6.4gt/s as reports 3200mhz in cpu-z. Slow mode for 200 fsb and higher?
|
|
|
Post by PolRoger on Jan 2, 2022 16:40:08 GMT -5
There are 4 QPI speed settings... Slow mode; 4.8GT/s=18x; 5.866GT/s= 22x; and 6.400GT/s = 24x. Different motherboard BIOS have different ways of reading QPI but CPUZ will read it in MHz. I have found that ~4000MHz QPI speed seems to be the upper limit in most cases. Ok, I'm running 6.4gt/s as reports 3200mhz in cpu-z. Slow mode for 200 fsb and higher? Uncore also has its own adjustable multi in BIOS separate from the CPU multi and the QPI multi. With 4.8GT/s = 18x the max is ~221 BCLK and going above that requires QPI slow mode.
|
|
|
Post by ShrimpBrime on Jan 2, 2022 16:43:39 GMT -5
Ok, I'm running 6.4gt/s as reports 3200mhz in cpu-z. Slow mode for 200 fsb and higher? Uncore also has its own adjustable multi in BIOS separate from the CPU multi and the QPI multi. With 4.8GT/s = 18x the max is ~221 BCLK and going above that requires QPI slow mode. That explains a lot of my troubles, thank you. I thought the qpi was the fsb issue I was having lol. Passed my quick run 4777mhz 8m6s 32m. Lopened Lowered clocks for the side challenge and running that at simple 133 bus clocks. See what it looks like.
|
|
|
Post by dr4g00n on Jan 2, 2022 16:46:42 GMT -5
Noticed the uncore on my evga classified boards are stuck at 20x for some reason? Never used to do that. Even flashed the bios and it's still doing it.
|
|
|
Post by PolRoger on Jan 2, 2022 16:52:22 GMT -5
Ok, I'm running 6.4gt/s as reports 3200mhz in cpu-z. Slow mode for 200 fsb and higher? Uncore also has its own adjustable multi in BIOS separate from the CPU multi and the QPI multi. With 4.8GT/s = 18x the max is ~221 BCLK and going above that requires QPI slow mode. For 5.866GT/s = 22x max is ~181 BCLK and for 6.400GT/s = 24x max is ~166 BCLK.
|
|
|
Post by ShrimpBrime on Jan 2, 2022 16:59:09 GMT -5
I want to see some efficiency comparisons What shall the terms be? I suggest 32M with a 4G CPU limit. We can have a side competition to see how the efficiency lines up. I will be running PSC and I am curious how far off it is from Hyper...I suspect pretty small difference. Here.
|
|
|
Post by ShrimpBrime on Jan 2, 2022 17:01:30 GMT -5
Noticed the uncore on my evga classified boards are stuck at 20x for some reason? Never used to do that. Even flashed the bios and it's still doing it. Some reason?? Some reason?? You overclocked the multiplier right off it's face hahah!!! XD
|
|
|
Post by ground on Jan 2, 2022 17:07:00 GMT -5
I want to see some efficiency comparisons What shall the terms be? I suggest 32M with a 4G CPU limit. We can have a side competition to see how the efficiency lines up. I will be running PSC and I am curious how far off it is from Hyper...I suspect pretty small difference. abload.de/img/842719wazaylj7w.png4GHz core, 4GHz uncore Hyper target. This is doable with half way decent Hypers on air. BDBG and other ICs I never managed anywhere near <2.1M at 4000/4000, no matter how tight. Cold PSC could come close or even beat hyper, but cold hypers are another level as well (6-6-4-18 at >2000) QPI clock itself doesn't have a big impact if at all. 32M, given a decent OS, doesn't lose time even in slowmode. Quickly benches (such as 1M, pifast and wprime 32M) benefit greatly from graphics drivers, but still run just fine in slowmode. There is a way to run pifast without graphics drivers, but I'm not gonna go into too much detail there. Most boards top out around 4000, a few such as some R3E and most R3BE can do a bit more, in the range of 4300-4500. Slow mode becomes necessery above 220 BCLK on most boards. To get over the 220 BCLK limit, you need to increase PCIe clock. A minimum ratio you'll want is roughly BCLK*0.45 = PCIe. Most boards will boot up to 118 PCIe, with setfsb 135 is typically doable. However, many CPUs are not up to high BCLK - most top out between 230 and 250 on ambient cooling, with unicorns (~1-2%) doing >270 on aircooling. Cold adds another 10-30MHz ontop of that. Uncore does a lot, also part of the reason a lot of old efficiency comps are beyond useless, since people ran uncore > core clocks. EVGA and MSI have a bug in the bios where setting uncore on 32nm xeons doesn't work usually. I hope I answered a couple questions that were going around.
|
|
|
Post by georgekokovinis on Jan 2, 2022 17:20:56 GMT -5
I want to see some efficiency comparisons What shall the terms be? I suggest 32M with a 4G CPU limit. We can have a side competition to see how the efficiency lines up. I will be running PSC and I am curious how far off it is from Hyper...I suspect pretty small difference. abload.de/img/842719wazaylj7w.png4GHz core, 4GHz uncore Hyper target. This is doable with half way decent Hypers on air. BDBG and other ICs I never managed anywhere near <2.1M at 4000/4000, no matter how tight. Cold PSC could come close or even beat hyper, but cold hypers are another level as well (6-6-4-18 at >2000) QPI clock itself doesn't have a big impact if at all. 32M, given a decent OS, doesn't lose time even in slowmode. Quickly benches (such as 1M, pifast and wprime 32M) benefit greatly from graphics drivers, but still run just fine in slowmode. There is a way to run pifast without graphics drivers, but I'm not gonna go into too much detail there. Most boards top out around 4000, a few such as some R3E and most R3BE can do a bit more, in the range of 4300-4500. Slow mode becomes necessery above 220 BCLK on most boards. To get over the 220 BCLK limit, you need to increase PCIe clock. A minimum ratio you'll want is roughly BCLK*0.45 = PCIe. Most boards will boot up to 118 PCIe, with setfsb 135 is typically doable. However, many CPUs are not up to high BCLK - most top out between 230 and 250 on ambient cooling, with unicorns (~1-2%) doing >270 on aircooling. Cold adds another 10-30MHz ontop of that. Uncore does a lot, also part of the reason a lot of old efficiency comps are beyond useless, since people ran uncore > core clocks. EVGA and MSI have a bug in the bios where setting uncore on 32nm xeons doesn't work usually. I hope I answered a couple questions that were going around. Extremely useful info. Thank you !
|
|
|
Post by ShrimpBrime on Jan 2, 2022 17:21:59 GMT -5
I want to see some efficiency comparisons What shall the terms be? I suggest 32M with a 4G CPU limit. We can have a side competition to see how the efficiency lines up. I will be running PSC and I am curious how far off it is from Hyper...I suspect pretty small difference. abload.de/img/842719wazaylj7w.png4GHz core, 4GHz uncore Hyper target. This is doable with half way decent Hypers on air. BDBG and other ICs I never managed anywhere near <2.1M at 4000/4000, no matter how tight. Cold PSC could come close or even beat hyper, but cold hypers are another level as well (6-6-4-18 at >2000) QPI clock itself doesn't have a big impact if at all. 32M, given a decent OS, doesn't lose time even in slowmode. Quickly benches (such as 1M, pifast and wprime 32M) benefit greatly from graphics drivers, but still run just fine in slowmode. There is a way to run pifast without graphics drivers, but I'm not gonna go into too much detail there. Most boards top out around 4000, a few such as some R3E and most R3BE can do a bit more, in the range of 4300-4500. Slow mode becomes necessery above 220 BCLK on most boards. To get over the 220 BCLK limit, you need to increase PCIe clock. A minimum ratio you'll want is roughly BCLK*0.45 = PCIe. Most boards will boot up to 118 PCIe, with setfsb 135 is typically doable. However, many CPUs are not up to high BCLK - most top out between 230 and 250 on ambient cooling, with unicorns (~1-2%) doing >270 on aircooling. Cold adds another 10-30MHz ontop of that. Uncore does a lot, also part of the reason a lot of old efficiency comps are beyond useless, since people ran uncore > core clocks. EVGA and MSI have a bug in the bios where setting uncore on 32nm xeons doesn't work usually. I hope I answered a couple questions that were going around. Holy sh!t. Let me get my jaw off the floor really quick..... 270! on Air! 990mhz+ CL what.... 6! So uncore I want, QPI I don't necc need, and try for Cas 6 at 1000mhz effective... And freeze everything. Gotcha. You're on another level. I'll keep poking at it though!
|
|
|
Post by dr4g00n on Jan 2, 2022 17:22:16 GMT -5
Maybe it's cpu dependant, in the past with a x5670 I was able to run whatever uncore I wanted but with my x5650 & x5687 I can't. And of course I went and cracked the die on said x5670 cleaning the die for LM.
|
|
|
Post by ShrimpBrime on Jan 2, 2022 17:24:01 GMT -5
Maybe it's cpu dependant, in the past with a x5670 I was able to run whatever uncore I wanted but with my x5650 & x5687 I can't. And of course I went and cracked the die on said x5670 cleaning the die for LM. Ho man that sucks. Sounds like a really good chip too.
|
|
|
Post by ground on Jan 2, 2022 17:30:58 GMT -5
|
|
|
Post by georgekokovinis on Jan 2, 2022 17:56:23 GMT -5
N.1 King of x58. Maybe Sam ( Tapakah ) is on the same level. Dont know. Congrats. BTW if you ever decide to sell one set of those mems, I am your man
|
|
|
Post by PolRoger on Jan 2, 2022 18:09:30 GMT -5
ground 's' experience and skill level with the X58 platform has given me the rabbit to chase after. I'll be working on improving/lowering my 4Ghz SuperPi 32M score on the side as well. A quick starting point run: EDIT: Working towards getting 4000MHz Uncore... I'm not sure if I have a mem kit that can run @2000c7?
|
|
|
Post by ShrimpBrime on Jan 2, 2022 22:37:51 GMT -5
ground 's' experience and skill level with the X58 platform has given me the rabbit to chase after. I'll be working on improving/lowering my 4Ghz SuperPi 32M score on the side as well. A quick starting point run: <button disabled="" class="c-attachment-insert--linked o-btn--sm">Attachment Deleted</button> EDIT: Working towards getting 4000MHz Uncore... I'm not sure if I have a mem kit that can run @2000c7? <button disabled="" class="c-attachment-insert--linked o-btn--sm">Attachment Deleted</button> Been at this for a minute, but I didn't take screen shots. For the following reason/s 143 bclk running 1000mhz CL 6-9-8-24 was fine. This required to run the Dominators in dual channel because I only have 2 sticks. Was able to run 1067mhz 8-9-8-24 + 1 bclk before I got the blue screens. None of this seemed to produce better figures over the mixed triple channel and looser timings. No matter, I did a ton of learning today. You guys are pretty hard core. I've never run 1000mhz Cas6 on any DDR3 platform, so you all helped accomplish it today. I'm greatful to OC with you guys. Thank you for all the help! Edit: I'll get you a screeny Cas 6 so you know I'm not BS'n. I've already went back to a different memory configuration. Elphida 1333 mems don't OC too high. I'd have to work on timings with these as a separate adventure, but I'm not digging them r/n so gonna pull em out too....
|
|